[NVBUG-6448152][test] TEST ONLY bisect native f4742f68

[chienchunhung]

TEST ONLY

Historical-source/current-runtime discriminator for a separate C++ throughput regression. This draft is diagnostic only and is not intended for merge.

• Native source checkpoint: f4742f6868, with exact native tree ebd8c9a8fdd239a2d40381e505985fb5b3445dc5.
• The checkpoint is an exact first-parent midpoint: 199 commits after the admission and bounded-polling change and 199 commits before the known-slow native checkpoint 544199a47b91b65d295322c6bb1f74dadbcf6a79.
• Runtime-only overlay: NIXL 1.3.1 and the current five CI image tags from build 202607151440-16194.
• The reviewed pre-cancellation asynchronous-consensus factor runs protocol mode 2 with cancellation fixed off.
• CMake context normalization and restoration have zero net semantic effect.
• No DSpark, workload YAML, selector, timeout, or product compatibility changes are included.
• The test target remains the exact GB300 3-node/12-GPU disaggregated performance selector. A result at or above 1402 output tokens/s places the regression in the later half; a result near 800 moves the slow bound to this checkpoint. Any build/import failure, timeout, protocol mismatch, cancellation, or failed request is non-interpretable.

The merged asynchronous-consensus change is not under re-review here; this draft only localizes the separate later throughput regression.

[chienchunhung]

/bot run --disable-fail-fast --stage-list "GB300-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge-1"

[tensorrt-cicd]

PR_Github #61747 [ run ] triggered by Bot. Commit: 9cf779e Link to invocation

[tensorrt-cicd]

PR_Github #61747 [ run ] completed with state FAILURE. Commit: 9cf779e
/LLM/main/L0_MergeRequest_PR pipeline #49949 (Partly Tested) completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[chienchunhung]

Terminal evidence for the native f4742f68 checkpoint at the exact TEST ONLY diagnostic head:

• The sole selected GB300 workload completed 512/512 requests with 0 failed requests in 5249.71 seconds.
• Official metrics: 798.96 output tokens/s, 0.10 requests/s, and 13582.30 total tokens/s.
• Context-transfer consensus protocol mode 2 (version 1 with cancellation disabled) was active on all four CTX PP ranks, ranks 0 through 3.
• CTX, GEN, and disaggregated applications shut down cleanly. No fatal, protocol-mismatch, double-free, or uncaught-exception evidence was found.
• The red result is solely the expected performance-regression gate after the valid workload completed. A post-run classifier noticed startup readiness noise, but only one workload was submitted and no retry ran; this evidence is not censored and should not be rerun.

The result is +0.55% versus the 794.61 current-source point and 51.58% of the historical 1548.84 point: no recovery. Therefore the following 199 first-parent commits through the later slow checkpoint are excluded as the origin, and the slow source bound moves to at or before this native checkpoint.